Artifact reduction for JPEG-compressed images with VQ and linear estimation

A new non-iterative algorithm to remove compression artifacts appearing in JPEG encoded images is proposed in this work. Since degradation caused by the transform and the quantization is difficult to describe mathematically, a clustering technique is used to analyze the statistics of the source and noise in the training. Then, coefficients of a linear prediction filter are pre-calculated and stored in a codebook. In the decoding stage, the linear filter with proper coefficients is applied to quantized transform coefficients and to pixels at block boundaries after the inverse DCT. Experimental results show that the proposed method can efficiently reduce coding artifacts with a low computational complexity